FinRiskAtlas - Evidence-Grounded Processing - Quantitative Reasoning: leaderboard
Metric: Numerical Accuracy (%). Source: arxiv.org. 33 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Ling-2.6-1T | 86.21 |
| 2 | GLM-5 | 84.48 |
| 3 | Nemotron 3 Ultra | 81.9 |
| 4 | Qwen 3 235B A22B | 81.03 |
| 5 | Qwen3-Next-80B | 80.17 |
| 6 | Qwen 3 32B | 79.66 |
| 7 | Seed 2.0 Mini | 79.31 |
| 8 | Seed 2.0 Lite | 79.31 |
| 9 | Qwen 3.7 Max | 78.45 |
| 10 | GLM-5.2 | 78.45 |
| 11 | GPT-5.6 Sol | 77.59 |
| 12 | Kimi K3 | 77.59 |
| 13 | DeepSeek V3 (0324) | 76.72 |
| 14 | GPT-5.4 Mini | 76.72 |
| 15 | GPT-5.6 Terra | 76.72 |
Interactive version: theaggregate.ai/benchmark?slug=finriskatlas-evidence-grounded-processing-quantitative-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-19.